Link to paper

The full paper is available here.

You can also find the paper on PapersWithCode here.

Abstract

Existence of a mechanism studied that weakly improves a policy objective on the distribution of student types to schools.
Mechanism must satisfy constrained efficiency, individual rationality, and strategy-proofness.
Mechanism need not exist in general.
New notion of discrete concavity (M$^{\natural}$-concavity) introduced.
Mechanism with desirable properties constructed when distributional objective satisfies M$^{\natural}$-concavity.
Several distributional objectives that are natural in the setting satisfy M$^{\natural}$-concavity.

John Maynard Keynes highlighted the combination of economic efficiency, social justice, and individual liberty as the political problem of mankind
Study this problem in a matching context between schools and students of different types
Social justice can be interpreted as the policy to improve a distributional objective on student types
Economic efficiency of a matching can be understood as the requirement that there exists no alternative matching that improves the welfare of students without undermining the distributional objective
Individual liberty corresponds to the property that no student is forced to be matched with a school against their will
Strategy-proofness is added as the fourth desideratum
Distributional policies are usually implemented by reserving seats or positions for target groups in society
Kamada and Kojima (2015, 2017, 2020) introduce and study matching markets with regional constraints
Hafalir et al. (2022) introduce the interdistrict school choice problem and study distributional policies
Suzuki et al. (2018) study a market without student types
Dur and Ünver (2019) study exchange programs
Delacrétaz et al. (2019) study refugee resettlement
Combe et al. (2022) study how to improve assignments upon the initial matching
Imamura and Kawase (2022) study Pareto efficient matchings
Yokote (2022) studies an object allocation problem with heterogeneous objects

Finite set of schools, students, and student types
Each student has a type and a preference relation over schools and outside option
Each school has a capacity
Matching assigns students to schools or outside option
Initial matching exists
Distribution vector for each school and type
Feasible distributions exist
Distributional objective function
Matching weakly improves objective if better than initial
Pareto dominates if better for at least one student
Constrained efficient if no better matching
Individual rationality if better than initial

There exists a mechanism that weakly improves the distributional objective and satisfies constrained efficiency, strategy-proofness, and individual rationality
For some distributional policies, there is no mechanism with the desirable properties
Example 1 establishes this fact
There are two matchings that weakly improve the distributional objective, satisfy constrained efficiency and individual rationality
If a mechanism satisfies the desired properties, then its outcome must be either matching µ or µ'
There is a profitable deviation for a student if they misreport their preferences
A new notion of discrete concavity is introduced
Pseudo M-concavity is defined
The failure of a mechanism with the desirable properties in Example 1 is attributed to the policy function failing pseudo M-concavity

Introduce a mechanism that achieves desirable properties when the distributional objective is pseudo M ♮-concave
Create a hypothetical matching market with school-type pairs, outside option, and students from the original market
Define preference order Ps over school-type pairs and the outside option
Define priority ordering of students that school-type pairs and (∅, ∅) use to rank students
Generalize the top trading cycles mechanism of Shapley and Scarf (1974)
Algorithm terminates when no student remains to be processed
Outcome is defined as the matching induced by the outcome of the hypothetical problem
Theorem 1: TTC weakly improves the distributional objective and satisfies constrained efficiency, individual rationality, and strategy-proofness

A matching µ satisfies the policy goal Ξ if the distribution associated with µ is in Ξ.
M ♮ -convexity is a weakening of a standard notion in the mathematics and operations research literatures.
TTC can be applied within the context of distributional goals.
For a policy function that is defined as the Chebyshev distance to an M ♮ -convex “ideal set” is pseudo M ♮ -concave.

Chebyshev distance is a metric used to measure the distance between two vectors.
A policy function is used to measure the minimum negative Chebyshev distance to a set of “ideal distributions” for diversity.
Theorem 3 states that the policy function is pseudo M ♮ -concave, and satisfies certain criteria.

Established general results based on pseudo M♮-concavity of the distributional objective
Applied results to a variety of distributional objectives
Exchange-feasibility policy requires total number of students assigned to a district’s school do not decrease
TTC weakly improves the distributional objective and satisfies constrained efficiency, individual rationality, and strategy-proofness
M♮-convexity of a set applicable for policy goals in the future

District-level diversity policies are considered instead of exchange-feasibility policies.
There are upper and lower bounds for each district and type.
The policy goal is not M♮-convex.

Studied principles of Keynes in a matching context
Challenge was to design a mechanism that achieves constrained efficiency, individual rationality, and strategy-proofness while improving upon a distributional objective
Identified a notion of discrete concavity called pseudo M ♮ -concavity
Provided an explicit algorithm that achieves those goals, TTC
Identified a general class of distributional objectives that satisfy sufficient conditions
TTC works by assigning students to school-type pairs based on a master priority list
TTC-M satisfies the policy goal, constrained efficiency, individual rationality, and strategy-proofness
Hypothetical problem with modified policy goal satisfies all conditions assumed by Suzuki et al.
TTC satisfies the policy goal, constrained efficiency, individual rationality, and strategy-proofness in the original problem